Market research

Market Research (Worksheet 2 support)

Gather cited, uncertainty-flagged market evidence to inform Worksheet 2 scoring. This skill collects evidence; it does not set scores — the team scores on Worksheet 2 using this evidence.

Three research tiers (capability-driven)

The skill supports three engines. First, check capability: is the Workflow tool available on this surface? That single check decides the default — not stakes, not surface guesswork.

IF Workflow is available (Claude Code): default to the workflow engines.

• Tier 3 (the plugin's own wf-market-research.js via scriptPath) applies to exactly two factors — Market volume and Economic viability; its number cross-checking is built for these.
• Every other factor, and every broad / open question, defaults to Tier 2 (built-in deep-research) — including Time to revenues, Implementation obstacles, External risks, competitor landscape and general context.

IF Workflow is NOT available (some Cowork / Agent-SDK surfaces): fall back to Tier 1, the bundled market-research subagent. Tier 1 is the fallback, not the everywhere default — use it only when the workflow engines genuinely do not exist. Never let a missing tool block the research.

The flags override the default when the team is explicit: --deep → Tier 2, --wtp → Tier 3 (both require Workflow; if it is missing, they degrade to Tier 1).

Tier 1 — Bundled subagent (fallback when Workflow is absent)

Spawn the bundled market-research agent via the Agent tool (subagent_type: "market-research"), with the opportunity, the specific factor/assumption, and constraints (geography, horizon, segment):

Agent({ subagent_type: "where-to-plai:market-research",
  prompt: "Research <factor> for the opportunity <opportunity>. Constraints: <segment / geography / horizon>. Return cited findings with 🔴/🟠/🟢 confidence and an open-questions list. Evidence only — no score." })

It works in an isolated context and returns a distilled, source-cited summary with confidence flags. For several independent questions, spawn one agent per question in parallel.

Tier 2 — Built-in deep-research workflow (broad rigor)

For broad, open questions, frame a market question and call it. Give the team a one-line qualitative heads-up, then proceed — e.g. "Running a broad deep-research pass that cross-checks sources and flags confidence; I'll write the cited evidence into the workspace." No confirm-gate, no cost/time estimate.

Workflow({ name: "deep-research", args: "<opportunity> — <factor>: <the specific market question, with segment/geography/horizon>" })

This is generic deep research (scope → parallel search → fetch → 3-vote verify → cited synthesis). Strong breadth and verification, but not WtP-tuned.

Tier 3 — Bundled WtP workflow (factor-tuned, numbers cross-checked)

For the quantitative factors, call the plugin's own workflow by scriptPath (ships with the plugin — no install step). Give the team a one-line qualitative heads-up, then proceed — e.g. "Running a WtP-tuned deep research pass that cross-checks every figure and flags confidence; I'll write the cited evidence into the workspace." No confirm-gate, no cost/time estimate.

Workflow({
  scriptPath: "${CLAUDE_PLUGIN_ROOT}/workflows/wf-market-research.js",
  args: { opportunity: "<name>", factor: "<one of the six factors>", constraints: "<segment / geography / horizon>" }
})

It scopes the factor into falsifiable sub-questions (top-down + bottom-up for sizing), fans out searches, then runs a 3-lens adversarial panel on every number (source-credibility / recency / scope-match, majority rule — downgrading confidence on figures that fail) and a completeness critic that flags what is still missing. It returns a distilled summary with 🔴/🟠/🟢 flags, ranges where sources disagree, and a prioritised open-questions list — evidence only, never a score.

The two workflows take different args shapes by design: Tier 2 (deep-research) takes a single string prompt; Tier 3 (wf-market-research) takes a structured object (opportunity / factor / constraints). Don't normalise one to the other.

What to do (any tier)

1. Frame the question. Identify the opportunity and which factor / assumption needs evidence. Pull open assumptions from where-to-play/assumptions.md when present, and research the riskiest one first — the assumption that, if proven wrong, would sink the opportunity (for a fresh map that is usually the compelling reason to buy).
2. Check capability, then pick the tier per the rules above (Workflow available → Tier 3 for Market volume / Economic viability, Tier 2 for broad questions; otherwise Tier 1). For a Tier 2/3 run, give the one-line heads-up and proceed.
3. Demand honesty about uncertainty. Distinguish evidenced from inferred or unknown, and researchable risk from irreducible uncertainty — there is no crystal ball, so name what more searching cannot settle rather than papering over it. Flag low-confidence claims rather than assert them. Treat any unsourced number as an assumption, not a fact.
4. Write evidence into the workspace. Append findings to the relevant where-to-play/02_attractiveness-map/<opportunity>.md factor section, each with its source link and a confidence colour (🔴 none / 🟠 some / 🟢 significant). Add still-open questions to where-to-play/assumptions.md.
5. Hand back. Summarise what is now evidenced vs. still assumed, and return to the attractiveness-map skill so the team can (re)score with this evidence.

Boundaries

• Evidence in, scores out by the team — never let research "decide" a factor, in any tier.
• Prefer primary and recent sources; cross-check figures across ≥2 independent sources before treating them as solid.
• Market figures drift and sources disagree — present ranges and disagreements honestly rather than a single confident number.

Additional resources

• references/research-guide.md — how to brief the engine per factor, source-quality and cross-checking rules, the confidence scheme, the "surface uncertainty, don't bluff" (I-CALM) principle, and tier-selection guidance.
• ${CLAUDE_PLUGIN_ROOT}/workflows/wf-market-research.js — the bundled WtP research workflow (Tier 3) source.